Voxel-based Bayesian lesion-symptom mapping.

Area of Science:

• Neuroscience
• Statistics
• Medical Imaging

Background:

• Voxel-based lesion-symptom mapping (LSM) commonly uses null hypothesis significance testing (NHST).
• NHST methods have limitations, including inability to infer no difference and requiring multiple-comparison correction.
• These limitations hinder precise lesion-symptom relationship analysis.

Purpose of the Study:

• To propose and evaluate a Bayesian approach for voxel-based lesion-symptom mapping.
• To address the limitations of NHST in LSM by offering clearer semantic inference.
• To eliminate the necessity for multiple-comparison correction in LSM.

Main Methods:

• Developed a Bayesian statistical model to directly analyze the posterior distribution of lesion-proportion differences.
• Applied the Bayesian method to simulated datasets for validation.
• Utilized the Bayesian method on patient data from a study on acute ischemic left-hemispheric stroke.

Main Results:

• The Bayesian approach provides inference with clearer semantics compared to NHST.
• The proposed method does not require multiple-comparison correction.
• Both simulated and real-world stroke data demonstrated the Bayesian approach's sensitivity in identifying regions associated with group differences.

Conclusions:

• The Bayesian approach offers a more robust and interpretable alternative to NHST for voxel-based lesion-symptom mapping.
• This method enhances the ability to detect specific brain regions linked to neurological symptoms.
• The findings support the utility of Bayesian inference in neuroimaging research for lesion analysis.